Every other team optimized the API call. We changed the shape of the work. Governance sets budgets. Observability watches everything. Analysts plan. Routers assign. Engineers execute. Runtime discipline keeps every run lean.
Layer 0–1 · Governance + Observability
Budget enforcement + cost tracking
Hard token cap: 8,000 total. Per-agent budgets enforced. Every call logged: agent, model, tokens, cost, latency. Kill switch fires if budget exceeded.
↓
Layer 2 · Concept-craft · T0 Scaffold
Scaffold Agent (qwen3.5-9b via Parasail)
Decomposes the query into 4 precise research tasks with context slices, done criteria, and token budgets per branch. Never does research itself.
↓ Layer 3 routing → cheapest model that can do the job
You.com Finance Research API → YC, Crunchbase, VC funding velocity
T1b Trend Mapper
Job market signals
You.com Search API → skill demand curves, posting velocity
T1c Futures Projector
Role emergence
You.com Research API + LlamaIndex RAG → Arc×Terrain framework
T1d Disruption Analyst
Displacement map
You.com Search API → which roles are displaced, transition windows
↓ Layer 5 runtime: compress each output to 300 chars before fan-in
T2 · Synthesis · premium model · one call
Curriculum Architect (DeepSeek V4 Flash via Parasail)
Receives only compressed structured JSON from 4 subagents — never sees raw search data. Synthesizes into 5 ranked job predictions with confidence scores + 4-phase adjustable curriculum.
Stack
${['You.com Search API','You.com Research API','You.com Finance Research API','Parasail qwen3.5-9b','Parasail DeepSeek V4 Flash','LlamaIndex RAG','Agno async','FastAPI','Render'].map(s=>`${s}`).join('')}